Cross-Domain Learning for Motor Fault Detection Using CORAL and Deep Networks

Abstract

Motor fault detection plays a vital role in industrial automation systems where operational continuity, safety, and efficiency are of paramount importance. Traditional fault detection methods often struggle to generalize across varying operational environments, which may differ in data distributions, sensor characteristics, or load conditions. To overcome this limitation, this study explores the application of cross-domain learning techniques, particularly CORrelation Alignment (CORAL), in conjunction with deep neural networks for robust motor fault diagnosis. By aligning feature distributions between source and target domains, CORAL minimizes domain shift, thereby enhancing the transferability of diagnostic models. A hybrid framework combining CORAL and convolutional neural networks (CNNs) is proposed, enabling effective feature extraction and domain-invariant representation learning. Experiments conducted using benchmark datasets from different motor operating conditions demonstrate significant improvements in fault classification accuracy, especially in target domains unseen during training. This paper provides a comprehensive evaluation of the proposed approach, highlighting its potential to facilitate scalable and adaptive fault detection in industrial systems.